When I was younger, I had a CD collection of which I was proud. We’re talking 1000+ albums. Yet after spending 15 years acquiring them, I moved across the country, to an apartment in which furniture like the sofa, the desk, the dining table only just fit. The CDs were going under the bed, and the case my dad had built was not coming with me.
And so began a project that lasted many nights after work: converting the content into MP3 files. When you undertake a project requiring that much effort, there is little point in half measures. In my case, that meant perfecting the metadata that automatically downloaded each time I slotted in a new disc. Artists and titles were, for example, only starting suggestions: they needed to be recast to sort by surname or without definite articles.
Still, applying those rules was easy; after all, there were rules. (Mostly. There were always exceptions. Does an act like Le Tigre fall under L or T?) It was genre classification that required the real thought, and I was fastidious, ending up with a taxonomy that had around sixty entries. Perhaps paradoxically, Alternative and Indie emerged as separate top-level categories, with twelve under the one, and eight under the other.
Subjectivity is always present in cataloging, and that may be especially true for something as personal and diverse in features as music. Tagging on music-focussed social websites, on which users can assign multiple and idiosyncratic genres to the same song, suggests this. In a sample retrieved from last.fm, for instance, Chen, Wright and Nejdl found “[o]n average, each track is associated with 29.9 tags” (2009). For my library, I could only choose one, and I wanted it to be right. Beyond pure organisation, my aim was enabling smarter playlist creation, and for a long time that was a useful approach.
But then came streaming. In particular, then came the ability to add content to my library, some of which I didn’t own, using multiple devices, some of which didn’t offer interfaces to read (much less amend) the files’ attributes. As iTunes morphed into Apple Music, and the separation between what I possessed and what I used blurred, the meticulousness of my data management began to slip. Slowly, over time, my music library was assimilated.
That has partly happened because of laziness. Although the single point of entry workflow is obsolete, the ability to refine the details after the fact still exists. But, at the same time, the details matter less now. Searching is richer, reducing the need for alphabetisation. For its part, playlist generation happens by pressing a button. Instead of aggregating songs I have somewhat reductively dubbed as Electronic/Italo-disco, I can rely on the ‘genius’ in the cloud, who isn’t limited to my personal collection, but has access to a catalog encompassing much of the music ever recorded.
As the intelligence on which we rely becomes increasingly artificial, information specialists will have to grapple with more complicated questions of how classification happens and what role subjectivity can or should play. Indeed, the Web 2.0 approach to genre assignment, studied in 2009, may already be obsolete. According to Panwar, et al, “[n]owadays precice [sic] tagging retrieval mechanisms are provided by deep feature extrating [sic] and learning methods. They attemp [sic] to understand almost all information inside a music from local to temporal features” (2017). That is more than a human can do. Indeed, Chen, et al., cite Weare, recounting “Microsoft required the assistance of 30 musicologists over a period of one year in order to manually label a ‘few hundred thousand songs.'” To the extent they don’t already, such approaches must surely apply to textual and other documents.
There’s a limit to how much one can realistically let go, though. Despite what iTunes said when I downloaded Oumou Sangaré’s most recent album, Africa is definitely not a genre.
No, every now and again, a person to take a stand.
Chen, L., Wright, P. and Nejdl, W. (2009) ‘Improving music genre classification using collaborative tagging data’, ACM International Conference on Web Search and Data Mining, Barcelona, Spain, 09-12 February 2009, viewed 9 October 2017, https://0-dl-acm-org.wam.city.ac.uk/citation.cfm?id=1498812 .
Panwar, S., Das, A., Roopaei, M. and Rad, P. (2017) ‘A deep learning approach for mapping music genres’, System of Systems Engineering Conference (SoSE), Waikoloa, HI, USA, 18-21 June 2017, viewed 9 October 2017, http://0-ieeexplore.ieee.org.wam.city.ac.uk/document/7994970/ .